The good (and popular) search engines use HTML structure.
Note that the hierarchical structure of HTML documents can be implied. This
is not as good as explicit structure using containment, but it is there.
> -- The ADDRESS element, which might seem to have useful semantics, has
> no internal structure. Wouldn't it be nice if a tool could extract
> an name and e-mail address and phone number from the address?
An excellent idea.